Convergence criterion

[Patrick Meyer]

I have two questions about the convergence criteiron in mirt. First, I am setting the convergence criterion to 0.001 using the call mod1<-mirt(data=dmirt, model=model, itemtype='3PL', TOL=0.001). Is this the correct call? 

Second, what exactly is the convergence criterion? Is it the change in likelihood or the maximum change in logits at each iteration or the maximum relative change in logits?

Thanks,

Patrick

[Phil Chalmers]

Hi Patrick,

On Sun, Jun 8, 2014 at 12:26 PM, Patrick Meyer <meye...@gmail.com> wrote:

I have two questions about the convergence criteiron in mirt. First, I am setting the convergence criterion to 0.001 using the call mod1<-mirt(data=dmirt, model=model, itemtype='3PL', TOL=0.001). Is this the correct call? 

Yes that appears fine.

 

Second, what exactly is the convergence criterion? Is it the change in likelihood or the maximum change in logits at each iteration or the maximum relative change in logits?

It is the maximum difference in all parameters after completing a full EM cycle. If the previous cycle had a vector of parameters 

last <- c(1,0,1,0)

and the next iteration had the values

next <- c(1.1, 0, 1.2, 0.1)

then the change would be max(abs(last - next)) == 0.2. The EM cycles stop when this value becomes less than TOL. Cheers.

Phil

 

Thanks,

Patrick

--
You received this message because you are subscribed to the Google Groups "mirt-package" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mirt-package...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Patrick Meyer]

That's what I thought, but I appreciate the clarification. Thanks!